New paper: Sea Ice Classification of SAR Imagery Based on Convolution Neural Networks

Congratulations to PhD candidate Salman Khalegian`s paper on Sea Ice Classification of SAR Imagery Based on Convolution Neural Networks.

Photos: Salman Khalegian (portrait) and Anthony Paul Doulgeris who capured scientists on board a reseach vessel that is navigating through sea ice covered waters in limited visibility.

Salman has been exploring new and existing neural networks for sea ice classification using Sentinel-1 synthetic aperture radar (SAR). The automatic analysis of sea ice in SAR images is challenging because of thermal noise effects and ambiguities in the radar backscatter from sea ice surfaces make it difficult for algorithms to detect the correct sea ice type.

To help solving this challenge, Salman and his colleagues use manually annotated SAR images containing various sea ice types, that are divided into small patches which are processed one at a time, to construct a dataset for Deep Learning (DL) analysis. The goal was investigating two key challenges: the absence and presence of sea ice or open-water, and the classification of sea ice types. The experimental results show both qualitatively and quantitatively that our models produce accurate classification results.

The paper can be accessed here.